Light evidence (grade D: anecdotal)The builder's own description and estimates, with no order or revenue figures.

A Claude Agent Takes Instagram DM Orders for a 7-Location Sushi Chain

For a delivery-focused sushi chain that gets about 90% of its orders through Instagram DMs, an agent now answers instantly, flags allergens, upsells and sends confirmed orders to the kitchen.

Who this is about: Delivery-focused sushi restaurant chain (7 locations)

ClaudeMeta API (Instagram DMs)The restaurant's CRMLocal & service Shopify & ecommerceAutopilot

Get the prompt

SalesAdded Oct 15, 2026Source: reddit.com

What it takes

Cost: PaidPay-as-you-go Claude API usage. The builder says prompt caching keeps it cheap.
Setup time: A few weeksOur estimate from the source, for one business.
Skill needed: Developer skillsYou build or run your own code, even if AI writes it.

Explain it simply

A sushi chain gets most of its orders as private messages on Instagram, and orders were lost when nobody answered fast. Now Claude, an AI, replies right away using the menu, allergy information and delivery areas, suggests extra items and sends confirmed orders to the kitchen. It works across all seven locations.

New to AI? See how AI works

What was done

The chain's revenue depended on one person typing DM replies, and orders were lost when messages were missed or answered late. The builder connected Instagram DMs to Claude with the chain's rules, tone of voice and a knowledge base covering menu, ingredients, calories, allergens, delivery zones, hours, prep times and promotions for all seven locations.

The agent helps customers choose, explains rolls, flags allergens, suggests add-ons, confirms the order, pushes it to the kitchen and logs it in the CRM and an owner dashboard.

How it went

The builder says about 97% of messages read the large menu block from Claude's prompt cache, which costs a tenth of the normal input price, so running a capable model on every DM stays cheap.

A secondary summary reports the builder's estimate that orders are right about 98% of the time. We couldn't find that figure in the original post.
Calls, voice notes, photos and requests for a human still go to staff.
Keeping menu and allergen data current is the main risk: a stale allergen entry is a safety problem, not just a wrong order.
The results are the builder's own estimates.

Prompt

PromptWritten by our editors from the source

Act as the order-taking agent for [restaurant], which has [number] locations.
Use the menu, ingredients, allergens, delivery zones, hours, prep times and
promotions below. Ask which location the customer wants, warn about allergens,
confirm the address is inside a delivery zone, suggest one add-on when it fits,
then read back the items and total and ask the customer to confirm.
Hand off to a human for voice notes, photos, calls or any request for a person.

Cost tip from the source: keep menu + rules in one stable cached prefix and
put only the live conversation after it.

Our version; the exact prompt wasn't shared. The caching tip is from the post.

Business fit

Who this helps, and how:

Retail store owners: take product questions and orders from Instagram or WhatsApp DMs after hours, with stock and prices in the knowledge base.
Local & service
Local service businesses: answer DM booking requests for cleaning, detailing or lawn care with service areas and prices preloaded.
Local & service
Startup founders and e-commerce sellers: handle pre-sale DM questions on sizing, ingredients or shipping, then hand off to checkout.
Shopify & ecommerce
Independent contractors (barbers, bartenders, consultants): let DMs book appointments and explain services while you work.
Local & service
Medical and health professional business owners: answer hours, location and service-menu questions in DMs, routing anything clinical to staff.
Local & service

Similar tools

Other tools for this job, and how they differ. Chosen by our editors, never paid for.

ManyChat
Instagram DM automation with AI replies, no coding needed.
Chatfuel
A no-code chatbot builder for Instagram and WhatsApp.
ChatGPT
An alternative model provider for a custom-coded DM agent.

“Light evidence” (grade D) means: A brief, early or in-progress claim. Treat it as an idea to test, not proof. How we grade. Spotted a mistake? Suggest a correction.